2404 11347

ISOTROPIC SURFACES AND MOMENT MAP FLOW
FRANÇOIS JAUBERTEAU AND YANN ROLLIN
Abstract. We consider the moduli space of isotropic maps from

a closed surface Σ to a symplectic affine space and construct a
arXiv:2404.11347v1 [math.DG] 17 Apr 2024
Kähler moment map geometry, on a space of differential forms on

Σ, such that the isotropic maps correspond to certain zeroes of
the moment map. The moment map geometry induces a modified
moment map flow, whose fixed point set correspond to isotropic
maps. This construction can be adapted to the polyhedral set-
ting. In particular, we prove that the polyhedral modified moment
map flow induces a strong deformation retraction from the space
of polyhedral maps onto the space of polyhedral isotropic maps.
1. Introduction
1.1. Motivations. Symplectic geometry is the natural mathematical fra-
mework for Hamiltonian mechanics. Gromov showed that certain sym-
plectic properties are flexible, thanks to convex integration, resulting in
h-principle theorems [8], whereas others are rigid, due to the existence
of pseudoholomorphic curves [9]. In particular the Gromov-Lees the-
orem [9, 13, 5] establishes an h-principle for the problem of isotropic
immersions in a symplectic manifold. As a corollary, it is possible to
approximate every immersed submanifold of an affine symplectic space
by isotropic immersed submanifolds, with respect to the C 0 topology.
Our research started as we were conducting some numerical experi-
ments for Lagrangian surfaces and their mean curvature flow ; numerical
simulations naturally involve piecewise linear geometry and we quickly
realized that the only explicit examples of closed piecewise linear La-
grangian were the polygons in C and some polygonal versions of the
Clifford torus in Cm . Assessing the state of the art, we noticed that
very little is known about piecewise linear symplectic geometry. Then,
we turned to the existence problem for piecewise linear isotropic sub-
manifolds in affine symplectic spaces [11, 17] and obtained the following
approximation theorem:
2010 Mathematics Subject Classification. Primary 5299, 53D12; Secondary

39A14, 39A70, 47B39, 53D50, 53D20, 53D30.
Key words and phrases. polyhedral maps, piecewise linear symplectic geometry,
Lagrangian tori, isotropic tori, isotropic immersions, Kähler moment map, moment
map flow.
Supported by IRL CRM-CNRS 3457, UQAM, CIRGET, UMR6629, ANR-21-
CE40-0017, CHL 11-LABX-0020-01.
1
2 FRANÇOIS JAUBERTEAU AND YANN ROLLIN
Theorem 1.1.1 ([17]). Let f : Σ # V be a smooth isotropic immersion,

where Σ is a surface diffeomorphic to a quotient torus, and (V, ωV ) is a
symplectic affine space. Then, there exists a family of piecewise linear
maps fN : Σ → V for N ∈ N, with the following properties:
(1) fN : Σ # V is a topological immersion;
(2) fN converges toward f in C 1 -norm.
As a corollary, every smoothly immersed torus in V can be approxi-
mated in C 0 -norm by isotropic piecewise linear immersed tori.
Surprisingly, the original proof of Theorem 1.1.1 relies on an infinite
dimensional Kähler moment map geometry and the fixed point princi-
ple developed in [11]. However, an alternate proof using only on soft
techniques was given subsequently by Etourneau [6], with significant
technical simplifications.
Several authors have been considering problems of piecewise linear
symplectic geometry, including Gratza [7] and Bertelson-Distexhe [3].
Panov also introduced a notion of polyhedral Kähler geometry in [16].
The space of piecewise linear symplectic maps of a 4-torus is also shown
to be related to a hyperKähler moment map geometry in [18]. It seems
that classical results of smooth symplectic differential geometry are cu-
riously challenging to generalize to the piecewise linear setting. Perhaps
this is a manifestation of symplectic rigidity ? To show how much has
to be accomplished in the field, here is a short list of challenging open
questions:
(1) The local Darboux theorem, well known in the smooth setting,
turns out to be a conjecture for piecewise linear symplectic man-
ifolds of dimension at least 4 (cf. [3] for a definition of piecewise
linear symplectic manifolds).
(2) The deformation theory of Lagrangian submanifolds is well de-
scribed in the smooth setting, thanks to the Lagrangian neigh-
borhood theorem [14]. However, there is no analogue of such a
deformation theory in the case of piecewise linear Lagrangian
submanifolds of a symplectic affine space. This motivates the
work and partial answers of [11, 17].
(3) The local structure of the infinite dimensional Lie group of sym-
plectic diffeomorphisms is well understood [2]. However, very
little is known about the local (and global) structure of the
space of piecewise linear symplectic maps, even in the case of a
4-dimensional torus [18].
The above questions seem hard settle for an obvious reason: the Moser’s
trick generally fails in the piecewise linear setting, although it may be
be used with significant efforts, in the case of piecewise linear volume
forms for instance [3] for instance. All these complications lead us to
speculate whether some exotica arise in the context of piecewise linear
symplectic geometry.
ISOTROPIC SURFACES AND MOMENT MAP FLOW 3
A Kähler moment map geometry inspired by Donaldson [4] was in-

troduced in [11] to construct the approximations fN of Theorem 1.1.1.
The construction starts from a Kähler surface (Σ, gΣ , JΣ , ωΣ ) and a
Hermitian affine space V with symplectic form ωV . These structures
induce a formal Kähler structure (M , G, J , Ω) with Kähler form Ω, on
the moduli space M = C ∞ (Σ, V ). The group of Hamiltonian diffeo-
morphisms Ham(Σ, ωΣ ) acts by precomposition on the moduli space
space M and preserves the formal Kähler structure. In fact the action
of Ham(Σ, ωΣ ) on M is formally Hamiltonian, with moment map given
by
f ∗ ωV
µD (f ) = − ,
ωΣ
understood as an element of the Lie algebra of Ham(Σ, ωΣ ), identified to
the space of smooth function C0∞ (Σ, R) orthogonal to constants. Thus,
zeroes of the moment map µD correspond to isotropic maps f ∈ M .
A detailed presentation of this moment map geometry is given in [11].
A finite dimensional approximation φN of the energy functional φ
of the moment map µD is considered in [11]. The downward gradient
flow of φN provides a flow of piecewise linear surfaces. A numerical
version of the flow and numerical experiments carried out in [11] pro-
vided some effective examples of piecewise linear Lagrangian tori in
C2 . Overall, this flow of piecewise linear surfaces has been interesting
from an experimental perspective. However, it is not satisfactory from
a mathematical point of view, for several reasons: we could not prove
that the flow is convergent, although it seems well behaved numerically.
Furthermore, the flow does not come from a finite dimensional moment
map picture and its geometrical interpretation is somewhat unclear.
As a response to these objections, a new moment map geometry
and its corresponding flow are introduced in this paper. They do not
have any of the above issues: the new moment map geometry can
be immediately adapted to the polyhedral setting and much stronger
mathematical results are obtained. In particular, Theorem E is a Duis-
termaat type theorem, which shows that the flow has the nicest possible
behavior.
1.2. Statement of results. Let V be a Hermitian affine space, with un-
derlying vector space V~ and and symplectic form ωV . We consider a
smooth closed oriented surface Σ, endowed with a Riemannian met-
ric gΣ , a compatible almost complex structure JΣ and a corresponding
Kähler form ωΣ = gΣ (JΣ , ·, ·).
The moduli space of smooth map f : Σ → V is denoted M . Formally
M is an affine space with Ω0 (Σ, V~ ) as the underlying vector space. The
differential of a smooth map f ∈ M defines an exact V~ -valued 1-form
df and we have a linear differential operator
d : M → F = Ω1 (Σ, V~ ).
The image of M by d is the space of exact 1-forms denoted F0 . We

show that the moduli space F carries a natural Euclidean L2 -inner
product G, defined by Formula (2.3), and a compatible almost complex
structure J , defined by Formula (2.5). The corresponding Kähler form
is denoted Ω = G(J ·, ·).
Formula (2.6) defines an action of the gauge group TC = C ∞ (Σ, C∗ )
on F . The action of the real subgroup T = C ∞ (Σ, R) is merely the
action by complex multiplication on V~ -valued differential 1-forms and
we have the following result:
Theorem A. Let Σ be a smooth closed surface endowed with a Kähler
structure (Σ, gΣ , JΣ , ωΣ ) and V , a Hermitian affine space.
The moduli space F = Ω1 (Σ, V~ ) carries a natural Kähler structure
(F , G, J , Ω) and an action of the gauge group TC = C ∞ (Σ, C∗ ).
The almost complex structure J is invariant under the action of
TC . The action of TC is the complexification of the T-action, where
T = C ∞ (Σ, S 1 ) is the real subgroup.
The T-action preserves the Kähler structure (F , J , G, Ω) and is Hamil-
tonian. The map µ : F → t, where t is the Lie algebra of T, given
by
F ∗ ωV
µ(F ) = − ,
ωΣ
is a moment map. In other words, µ is T-invariant and, for every
ζ ∈ t,
Dhhµ, ζii = −ιXζ Ω,
where Xζ is the vector field on F defined by the infinitesimal action of
ζ on F .
Corollary B. With the assumptions of Theorem A, a smooth map f :
Σ → V is isotropic if, and only if,
µ(df ) = 0.
Remark 1.2.1. In the finite dimensional setting, the famous Kempf-
Ness theorem relates the existence of zeroes of a Kähler moment map in
a complexified orbit with an algebro-geometric notion of stability [15].
In view Theorem A and Corollary B, it is tempting to try to extend this
theory in our case: the existence of isotropic maps should be related to
some kind of algebraic condition. The answer turns out to be somewhat
trivial for the total moduli space F , as discussed at §2.12. However,
we are mainly interested in the zeroes of the moment map that belong
to the subspace F0 ⊂ F , as they are related to isotropic maps by
Corollary B. Unfortunately, F0 is not invariant under the gauge group
action and it is not clear how to adapt the classical theory from this
point.
We consider the energy functional of the moment map

φ:F →R
given by
1
φ(F ) = kµ(F )k2L2 .
2
By construction, φ is non negative and the vanishing locus of φ agrees
with the vanishing locus of the moment map. Our idea is to interpret
the functional φ as a Morse-Bott function on the moduli space F ,
in the spirit of Atiyah-Bott [1], who considered the case of the Yang-
Mills functional. The expectation is that the flow is going to produce a
strong deformation retraction of M onto the space of isotropic maps.
However, it is not clear whether φ satisfies the Morse-Bott condition
in any reasonable sense. Another difficulty is that the subspace F0
is not gauge invariant and may not be preserved by the usual Morse-
Bott gradient flow. We get around this issue by defining the modified
moment map flow, as the downward gradient flow of the restricted
functional φ : F0 → R. More precisely, the modified moment map flow
along F0 is given by the evolution equation
∂F
= −∇◦ φ(F ),
∂t
where F ∈ F0 and ∇◦ φ is the gradient of the restricted functional.
Our first theorem shows that the flow is geometrically relevant and is
a well posed problem from an analytical point of view:
Theorem C. The fixed point locus of the modified moment map flow
on F0 , in other words the critical set of φ : F0 → R, agrees with the
vanishing locus of φ : F0 → R. Furthermore,
(1) the modified moment maps flow has the short time existence
property and
(2) the L2 -norm is non increasing along the flow.
Remarks 1.2.2. (1) The definition of the short time existence prop-
erty involves the use of Hölder completions of the moduli space F .
A technical version of the short time existence is given at The-
orem 3.1.3.
(2) Unfortunately, the modified moment map flow does not seem
to have any nice regularizing properties, like parabolic flows for
instance. Consequently, the long time existence of the flow is
unclear.
(3) In the case where dimR V = 4, we show at Corollary 3.2.3 that
φ : F0 → R is a Morse-Bott function, in a neighborhood of
every monomorphism (cf. Definition 2.10.2) of its vanishing
locus.
The purpose of §4 is to extend all the above constructions to the

polyhedral setting. The concepts needed for stating the results are
quickly introduced below and the reader may refer directly to §4 for
the exact definitions.
A polyhedral surface Σ is a topological surface endowed with a tri-
angulation T and a polyhedral metric gΣ . An orientation of Σ induces
a polyhedral Kähler form ωΣ and a polyhedral almost complex structure
JΣ adapted to gΣ .
A polyhedral map with respect to a triangulation T is a continuous
map f : Σ → V such that the restriction of f to every simplex of the
triangulation is an affine map. The moduli space of polyhedral maps
f : Σ → V is denoted M (T ).
A polyhedral map is generally not differentiable. However, the re-
striction of a polyhedral map to any simplex of the triangulation has a
well defined tangent map. Accordingly, a polyhedral map f : Σ → V is
called a polyhedral isotropic map, if the pullback of ωV by f , restricted
to every simplex of the triangulation, vanishes identically.
The existence of polyhedral isotropic immersions in M (T ) is an
open question. The proof of Theorem 1.1.1 proceeds by introducing a
particular sequence of triangulations TN , in the case of the 2-torus Σ,
with a large number of simplices of order O(N 2 ) and stepsize of order
O(N −1 ). The piecewise linear approximations fN are in fact polyhedral
isotropic maps that belong to M (TN ). However, the topology of the
space of polyhedral isotropic immersions in M (T ) remains completely
mysterious for a fixed triangulation T of Σ.
Returning to the general case of a oriented polyhedral surface Σ, we
pursue the analogy with the smooth setting. The space vector space
F (T ) is the space of families F = (Fσ )K2 , where σ belongs to the set of
facets K2 of the triangulation T and Fσ is a constant V~ -valued 1-form
on σ. Because the restriction of a polyhedral map to every simplex of
the triangulation is differentiable, there is a natural differential map
d : M (T ) → F (T )
and its image is denoted F0 (T ).
We also define the group TC (T ) as the space of families λ = (λσ )σ∈K2 ,
where λσ : σ → C∗ is a constant function. The real subgroup of families
λ with λσ ∈ S 1 is denoted T(T ). The group TC (T ) acts on F (T )
by Formula (4.2). As in the smooth case, we construct a Euclidean
inner product G, a compatible almost complex structure J and a cor-
responding Kähler form Ω on F (T ). Thus we have a Kähler structure
(F (T ), G, J , Ω) with an action of TC (T ) and we can state our next
result:
Theorem D. Theorem A and Corollary B hold in the polyhedral setting.
As in the smooth setting, we define an energy of the moment map
φ : F (T ) → R by φ(F ) = 21 kµ(F )k2L2 . The downward gradient of
the functional φ restricted to F0 (T ) is called the polyhedral modified

moment map flow.
The moduli spaces M (T ) and F (T ) are finite dimensional and
much stronger result than Theorem C are expected, as the polyhedral
moment map flow is an ordinary differential equation. This is indeed
the case, and we obtain the following Duistermaat type theorem:
Theorem E. For F ∈ F0 (T ), the polyhedral modified moment map

flow admits a unique solution Ft ∈ F0 (T ), defined for t ∈ [0, +∞),
such that F0 = F . Furthermore, Ft admits a limit F∞ ∈ F0 (T ) as t
goes to +∞, with the property that φ(F∞ ) = 0.
The extended flow
Θ : [0, +∞] × F0 (T ) → F0 (T )
defined by Θ(t, F ) = Ft for t ∈ [0, +∞) and Θ(+∞, F ) = F∞ has the

following properties:
(1) The fixed point locus of the flow Θ is the vanishing set of φ :
F0 (T ) → R.
(2) The flow defines a strong deformation retraction of F0 (T ) onto
the vanishing locus of φ : F0 (T ) → R.
(3) There are non trivial flow lines converging toward 0. More pre-
cisely, there exists F ∈ F0 (T ) \ 0, such that Θ(+∞, F ) = 0.
(4) The flow Θ has exponential convergence rate in a neighborhood
of every regular point of the vanishing locus of φ : F0 (T ) → R.
The flow Θ on F0 can be lifted as a flow Θ̂ on M (T ) and we have

the following corollary:
Corollary F. There exists a unique map
Θ̂ : [0, +∞] × M (T ) → M (T )
such that
(1) d ◦ Θ̂t = Θt ◦ d and
(2) Θ̂0 = id on M (T ),
where we used the notation Θ̂t = Θ̂(t, ·). The map Θ̂ is the flow of the
evolution equation given by Formula (4.4) and defines a strong defor-
mation retraction of M (T ) onto the subspace of polyhedral isotropic
maps.
The existence of regular points (cf. Definition 4.10.1) of the vanishing

set of φ : F0 (T ) → R is not well understood in comparison to the
smooth setting (cf. §3.2). However, we show that the approximating
scheme of [11, 17] provides examples of regular points:
Theorem G. In the case where Σ is a 2-torus and f : Σ # V is

a smooth isotropic immersion, let fN ∈ M (TN ) be the Jauberteau-
Rollin-Tapie isotropic polyhedral immersions of Theorem 1.1.1 approxi-
mating f . Then FN = dfN ∈ F0 (TN ) is a regular point of the vanishing
locus of φ : F0 (TN ) → R for every sufficiently large N.
1.3. Open problems and future research. In view of the approxima-
tion scheme given by Theorem 1.1.1, it seems sensible to expect that
some type Gromov-Lees theorem should apply to the polyhedral set-
ting, provided some flexibility for the choice of triangulation T of Σ.
If this is indeed the case, we expect the following consequence:
Conjecture H. Let Σ be a surface endowed with a triangulation T .
Then, up to passing to a subdivision of T , the space of polyhedral
isotropic immersions in M (T ) is homotopically equivalent to the space
of smooth isotropic immersions of Σ in V .
For triangulation with lower complexity, an adapted Morse-Bott co-
homology theory for φ : F0 (T ) → R should lead to topological in-
variants for the space of non constant polyhedral isotropic maps. The-
orem E shows that the polyhedral modified moment map flow is ex-
tremely well behaved, which is a strong incentive to develop a Morse-
Bott theory, with a renormalized flow briefly discussed at §4.11.
A numerical version of the polyhedral modified moment map flow
is currently being developed for testing purposes of the Morse-Bott
theory and for producing effective examples of polyhedral isotropic im-
mersed surfaces. We sketch the relevant mathematical ingredients of a
computer program that produces approximate solutions of the flow at
§4.12. The code is to be released very soon [10].
1.4. Acknowledgments. A part of this research was carried out in 2023-
2024, thanks to the support of the CNRS International Research Lab
CRM in Montreal. Yann Rollin was hosted by the CIRGET at UQAM
and wishes to thank all his colleagues for creating such an enjoyable and
stimulating mathematical environment, in particular, Antonio Alfieri,
Vestislav Apostolov, Steven Boyer, Charles Cifarelli, Olivier Collin,
Alexandra Haedrich, Julien Keller, Abdellah Lahdili, Steven Lu, Dun-
can McCoy, Frédéric Rochon and Carlo Scarpa. We also thank Mélanie
Bertelson and Vincent Borrelli, for some useful discussions.
Contents
1. Introduction 1
1.1. Motivations 1
1.2. Statement of results 3
1.3. Open problems and future research 8
1.4. Acknowledgments 8
2. Kähler moment map 9
2.1. Target space 9

2.2. Source space 10
2.3. Fiberwise structures on tensor bundles 10
2.4. Moduli spaces and differentials 11
2.5. Euclidean structure and Hodge theory 11
2.6. Kähler structures on the moduli space 13
2.7. An involution 13
2.8. Gauge group action 14
2.9. Infinitesimal gauge group action 15
2.10. Symplectic density 15
2.11. Hamiltonian action 17
2.12. Stability and isotropic maps 17
2.13. Energy of the moment map 18
3. The modified moment map flow 19
3.1. Basic properties of the flow 20
3.2. Energy as a Morse-Bott function 21
4. Polyhedral isotropic surfaces 22
4.1. Definition of polyhedral surfaces 22
4.2. Whitney cohomology 23
4.3. Orientation and Kähler structure 23
4.4. Polyhedral maps and differentials 24
4.5. Moduli space structure in the polyhedral setting 25
4.6. Polyhedral symplectic density 26
4.7. Hamiltonian gauge group action 27
4.8. Polyhedral modified moment map flow 28
4.9. A Duistermaat theorem 29
4.10. Regular points of the moduli space 30
4.11. Topology of the moduli space 32
4.12. Numerical flow 34
References 37
2. Kähler moment map

This section is devoted to the description of an infinite dimensional
moment map geometry, which provides an interpretation of smooth
isotropic maps as some particular zeroes of a moment map.
2.1. Target space. Let V be an affine space and V~ , its underlying

vector space. We assume that V~ is a complex vector space, of complex
dimension m ≥ 2, endowed with a Hermitian inner product hV , anti-
C-linear in the first variable. For every v1 , v2 ∈ V~ , the decomposition
hV (v1 , v2 ) = gV (v1 , v2 ) + iωV (v1 , v2 ),

into the real and imaginary parts of hV provides a Euclidean inner

product gV and a symplectic form ωV . Multiplication by i defines a lin-
ear endomorphism i : V~ → V~ , also called an almost complex structure.
By definition the almost complex structure i is compatible with gV and
ωV , in the sense that
gV (iv1 , iv2 ) = gV (v2 , v2 ) and ωV (v1 , v2 ) = gV (iv1 , v2 )
for every v1 , v2 ∈ V~ .
2.2. Source space. Let Σ be a smooth closed and oriented surface,

endowed with a Riemannian metric gΣ . We also assume that Σ is con-
nected in all this paper, for simplicity of notations. We denote by ωΣ the
volume form of gΣ compatible with the orientation. The correspond-
ing almost complex structure JΣ ∈ End(T Σ) is defined as a fiberwise
rotation of T Σ → Σ with angle + π2 according to the orientation. By
construction, JΣ is compatible with gΣ and ωΣ is the associated Kähler
form, in the sense that
gΣ (JΣ η1 , JΣ η2 ) = gΣ (η1 , η2 ) and ωΣ (η1 , η2 ) = gΣ (JΣ η1 , η2 ),
for every x ∈ Σ and η1 , η2 ∈ Tx Σ. In conclusion, Σ is endowed with a
Kähler structure
(Σ, gΣ , JΣ , ωΣ ).
2.3. Fiberwise structures on tensor bundles. The bundle T ∗ Σ → Σ

is identified to T Σ → Σ using the duality induced by the Riemannian
metric gΣ . Thus T ∗ Σ → Σ and all the tensor bundles are endowed with
an induced fiberwise Euclidean inner product denoted gΣ as well.
The structure of complex vector space on V~ induces a canonical
structure of complex vector bundle on Λn Σ ⊗ V~ → Σ, where the tensor
product is taken with respect to R. Furthermore, hV and gΣ induce a
fiberwise Hermitian inner product on Λn Σ ⊗ V~ → Σ, denoted h, and
defined by the following property: for every x ∈ Σ, β1 , β2 ∈ Λnx Σ and
v1 , v2 ∈ V~ , then
h(β1 ⊗ v1 , β2 ⊗ v2 ) = gΣ (β1 , β2 )hV (v1 , v2 ).
The complex structure JΣ acts on Λn Σ → Σ, by composition on the
right: for every x ∈ Σ, β ∈ Λnx Σ and η1 , · · · , ηn ∈ Tx Σ, we put
(β ◦ JΣ )(η1 , · · · , ηn ) = β(JΣ η1 , · · · , JΣ ηn ).
This action is isometric, in the sense that for every β1 , β2 ∈ Λnx Σ, we
have gΣ (β1 , β2 ) = gΣ (β1 ◦ JΣ , β2 ◦ JΣ )
The action of JΣ extends canonically to Λn Σ ⊗ V~ → Σ and the
Hermitian product h is JΣ -invariant, in the sense that, for every β1 , β2 ∈
Λnx Σ ⊗ V~ , we have
h(β1 , β2 ) = h(β1 ◦ JΣ , β2 ◦ JΣ ).
We consider the Riemannian metric g given by the real part of h.

Then JΣ and i act isometrically on Λn Σ ⊗ V~ → Σ, in the sense that
g(β1 , β2 ) = g(β1 ◦ JΣ , β2 ◦ JΣ ),
g(iβ1 , iβ2 ) = g(β1 , β2 ).
We define a fiberwise almost complex structure on T ∗ Σ ⊗ V~ → Σ, by
the formula
J · F = −F ◦ JΣ . (2.1)
By definition, for every F1 , F2 ∈ Tx Σ ⊗ V~ , we have
∗
g(iF1 , iF2 ) = g(J · F1 , J · F2 ) = g(F1 , F2 ).

In particular, the formula
ω(F1 , F2 ) = g(J · F1 , F2 ) (2.2)
defines a fiberwise symplectic form on the bundle T ∗ Σ ⊗ V~ → Σ.
2.4. Moduli spaces and differentials. We consider the moduli space of
smooth V~ -valued differential 1-forms
F = Ω1 (Σ, V~ )
and the moduli space of smooth maps
M = C ∞ (Σ, V ).
Notice that M is an affine space with Ω0 (Σ, V~ ) as the underlying vector
space. For every smooth map, f : M → V , the tangent map f∗ : T Σ →
T V = V × V~ can be regarded as a differential
df : T Σ → V~
given by df = π2 ◦ f∗ , where π2 : T V → V~ is the second canonical pro-
jection. Hence we have a differential operator between moduli spaces
d
M −→ F .
The image of d is the subspace of exact V~ -valued differential forms,
denoted F0 ⊂ F . Furthermore, d is injective up to translations by
constant map in Ω0 (Σ, V~ ), identified to V~ . Thus, d induces a bijection
d
M /V~ −→ F0 .
2.5. Euclidean structure and Hodge theory. The fiberwise Euclidean
inner product g on the vector bundle Λn Σ ⊗ V~ → Σ induces an L2 -
Euclidean inner product on Ωn (Σ, V~ ), given by
Z
G(β1 , β2 ) = g(β1 , β2 ) ωΣ , for β1 , β2 ∈ Ωn (Σ, V~ ). (2.3)
Σ
For simplicity, we will often use the notations
hβ1 , β2 i = g(β1 , β2 ), hhβ1 , β2 ii = G(β1 , β2 )
and p p
|β| = g(β, β), kβkL2 = G(β, β).
The formal adjoint d⋆ of d is defined by the property

G(β1 , dβ2 ) = G(d⋆ β1 , β2 ) for β1 ∈ Ωn+1 (Σ, V~ ) and β2 ∈ Ωn (Σ, V~ )
and the Laplace operator ∆ is given by the formula
∆ = dd⋆ + d⋆ d.
The classical Hodge theory extends to V~ -valued differential forms. In
particular, we obtain a G-orthogonal projection onto the space of V~ -
valued exact 1-forms denoted
Π : F → F0 .
Hölder spaces are better suited for elliptic operators, as the Laplace
operator ∆. Recall that the Riemannian metric gΣ and the fiberwise
inner product g induce a C k,ν -Hölder norm for tensor fields over Σ,
denoted k · kk,ν , where k ∈ N is the number of derivatives and ν ∈ (0, 1)
is the Hölder regularity exponent for the k-th derivative (cf. [12] for an
explicit definition).
Hölder norms define Hölder completed spaces. In particular, we
denote by F k,ν and M k,ν the completion of F and M . For every
F ∈ F k,ν with k ≥ 0, the Hodge decomposition theorem states that F
admits a G-orthogonal decomposition
F = Fh + ∆G,
where G ∈ F k+2,ν and Fh is a smooth harmonic form. In particular,
we can define an orthogonal Hodge projection Π : F k,ν → F0k,ν by
Π(F ) = dd⋆ G
of F onto its exact component. By Proposition 2.5.1, the projector
Π is continuous with respect to Hölder topology. This result, which
is an immediate consequence of Hodge theory, is a crucial argument
for the proof of Theorem C, via the Cauchy-Lipschitz theorem (cf.
Theorem 3.1.3). We provide a proof of the proposition for the sake of
self-containedness.
Proposition 2.5.1. For k ≥ 0 and ν ∈ (0, 1), the orthogonal projection
Π : F k,ν → F0k,ν onto the exact component of a differential form is a
continuous linear map with respect to the C k,ν -norm. In other words,
there exists a real constant c > 0 such that
kΠ(F )kk,ν ≤ ckF kk,ν , for every F ∈ F k,ν .
Proof. For F ∈ F k,ν , the Hodge decomposition theorem shows that
F = Fh + ∆G, (2.4)
where G ∈ F k+2,ν is a 1-form orthogonal to harmonic forms and Fh is
the harmonic part of F .
A C k,ν -estimate for F provides a control on the C k−2,ν -norm of ∆F .
By Formula (2.4), we have ∆F = ∆2 G, since Fh is harmonic. Hence
the C k,ν -norm of F control the C k−2,ν -norm of ∆2 G. The operator ∆

is selfajoint, hence ∆G is orthogonal to the kernel of ∆. Then, elliptic
Schauder estimates provide a control on the C k,ν -norm of ∆G. Since G
was chosen orthogonal to harmonic forms, we deduce a C k+2,α control
on G by the Schauder estimates. In conclusion, there exists a universal
constant c1 > 0, such that
kGkk+2,ν ≤ c1 kF kk,ν .
Finally, the C k+2,ν -norm of G controls the C k,ν -norm of dd⋆ G and we
conclude that there exists a universal constant c > 0 that satisfies the
proposition.
2.6. Kähler structures on the moduli space. The space of differential
forms Ωn (Σ, V~ ) has a structure of module over C ∞ (Σ, C), acting by
complex multiplication on V~ -valued forms. More precisely
(λF )x = λ(x)Fx
for every λ ∈ C ∞ (Σ, C), F ∈ Ωn (Σ, V~ ) and x ∈ Σ.
The space of differential 1-forms F admits an alternate almost com-
plex structure
J : F → F,
defined by
(J · F )x · η = (JFx ) · η = −Fx ◦ JΣ · η, (2.5)
where J is defined by Formula (2.1), for every x ∈ Σ and η ∈ Tx Σ. By
construction, the almost complex structure J is compatible with the
Eulidean L2 -inner product G on F . The corresponding Kähler form Ω
deduced from G and J is given by
Ω(Ḟ1 , Ḟ2 ) = G(J Ḟ1 , F2 )
for every Ḟ1 , Ḟ2 ∈ F . Equivalently
Z
Ω(Ḟ1 , Ḟ2 ) = ω(Ḟ1 , Ḟ2 )ωΣ ,
Σ
where ω(Ḟ1 , Ḟ2 ) = g(J · Ḟ1 , Ḟ2 ).

In conclusion, we have a natural Kähler structure
(F , G, J , Ω)
on the moduli space F .
2.7. An involution. The complex vector space F admits several almost
complex structures: the almost complex structure i : F → F , corre-
sponding to the multiplication by i and the almost complex structure
J described above. By construction, the two almost complex structure
i and J commute. Therefore, the endomorphism
R:F →F
defined by
RF = iJ · F
is a linear isometric involution of F , which commutes with i and J .
We obtain a G-orthogonal decomposition
F = F+ ⊕ F−
where F ± are the eigenspaces associated to the eigenvalues ±1 of R.
More explicitely, we have
F + = {F ∈ F , F ◦ JΣ = iF } and
F − = {F ∈ F , F ◦ JΣ = −iF }.
In other words, F + (resp. F − ) is the subspace of complex (resp. anti-
complex) morphisms F : T Σ → V~ . Hence, every F ∈ F admits a
unique orthogonal decomposition
F = F + + F −,
where F ± ∈ F ± . By definition, we have
RF = F + − F − .
2.8. Gauge group action. We consider the infinite dimensional com-

plex Lie group
TC = C ∞ (Σ, C∗ ),
with trivial Lie algebra
tC = C ∞ (Σ, C).
We define an action of TC on F as follows: given λ ∈ TC and F ∈ F ,
we put
λ · F = λ̄−1 F + + λF − , (2.6)
where F = F + + F − according to the splitting F = F + ⊕ F − and
λ̄ denotes the complex conjugate. By construction the action of TC
preserves the almost complex structure J for the following obvious
reason:
Lemma 2.8.1. Each subspace F ± is J invariant and the action of TC
on F is J -linear.
The group TC contains the real subgroup
T = C ∞ (Σ, S 1 ),
where S 1 ⊂ C is the unit circle. If λ ∈ T, then λ̄−1 = λ, and group
action is merely given by complex multiplication:
λ · F = λF, for every λ ∈ T.
2.9. Infinitesimal gauge group action. For ζ ∈ tC = C ∞ (Σ, C) we

define an exponential map
exp : tC → TC ,
by
exp(ζ) = eiζ ,
so that the space of real valued functions t = C ∞ (Σ, R) is identified to
the Lie algebra of T. As usual, the infinitesimal action of ζ ∈ tC is the
vector field Xζ on the moduli space F , defined by
d
Xζ (F ) = exp(tζ) · F.
dt t=0
Formula (2.6) shows that

Xζ (F ) = iζ̄F + + iζF − . (2.7)
In particular, if ζ ∈ t, we have
Xζ (F ) = iζF. (2.8)
If, on the contrary, ζ ∈ it is a purely imaginary function, we have
Xζ (F ) = −iζRF. (2.9)
By (2.8) and (2.9), the infinitesimal action of ζ ∈ t satisfies
J · Xζ (F ) = J · iζF = ζRF = −i(iζ)RF = Xiζ (F ),
and we deduce the following result:
Lemma 2.9.1. The action of TC on F is the J -complexification of the
action of T.
In addition, the gauge group T acts isometrically on F . Indeed for
every λ ∈ T and F ∈ F ,
kλ · F k2L2 = kλF k2L2 = kF k2L2 .
In conclusion, we have the following result:
Proposition 2.9.2. The action of T on F is linear and preserves the
Kähler structure (F , G, J , Ω).
2.10. Symplectic density. The map
µ:F t = C ∞ (Σ, R)
(2.10)
F − 21 g(F, RF ).
can be interpreted as a symplectic density, according to the following
lemma:
Lemma 2.10.1. The following formulas hold

1 1 + 2 − 2 F ∗ ωV
µ(F ) = − ω(iF, F ) = − (|F | − |F | ) = − , (2.11)
2 2 ωΣ
for every F ∈ F , where the pullback is defined by (F ∗ ωV )(η1 , η2 ) =
ωV (F (η1 ), F (η2 )), for every x ∈ Σ and η1 , η2 ∈ Tx Σ.
Proof. Using the fact that − 21 g(RF, F ) = − 12 g(iJ F, F ) = − 12 g(J iF, F ) =
− 21 ω(iF, F ) we deduce the first identity.
Using the decomposition F = F + + F − , we write g(RF, F ) =
g(F + − F − , F + + F − ). Using the fact the F + and F − are pointwise
g-orthogonal, we deduce that g(RF, F ) = |F + |2 − |F − |2 , which proves
the second identity.
Let x ∈ Σ and (e1 , e2 ) be a gΣ -orthonormal oriented basis of Tx Σ.
In particular ωΣ (e1 , e2 ) = 1 and JΣ e1 = e2 . Hence (J · F )(e1 ) =
−F (JΣ e1 ) = −F (e2 ). Similarly (J · F )(e2 ) = F (e1 ). Hence RF (e1 ) =
−iF (e2 ) and RF (e2 ) = iF (e1 ). By definition
1
µ(F )(x) = − g(RF, F )(x)
2
1 1
= − gV (RF (e1 ), F (e1 )) − gV (RF (e2 ), F (e2 ))
2 2
1
= − (−gV (iF (e2 ), F (e1 )) + gV (iF (e1 ), F (e2 )))
2
= −ωV (F (e1 ), F (e2 ))
(F ∗ ωV )(e1 , e2 )
=− ,
ωΣ (e1 , e2 )
which proves the last identity of the lemma.
We introduce some h-principle terminology before stating an imme-
diate corollary below.
Definition 2.10.2. A differential form F is called isotropic if F : T Σ →
V~ maps every tangent space to an isotropic subspace of V or, equiva-
lently, if F ∗ ωV = 0. If the restriction of F : T Σ → V~ to every tangent
space is injective, we say that F is a monomorphism.
Corollary 2.10.3. Let F be an element of F . The following properties
are equivalent:
(1) F ∈ F is isotropic
(2) µ(F ) = 0 ∈ t .
(3) |F + | = |F − | identically on Σ.
If F ∈ F0 , we deduce the following interpretation for isotropic maps:
Corollary 2.10.4. For every map f ∈ M , the following properties are
equivalent:
(1) f is an isotropic map.
(2) F = df ∈ F0 is isotropic.
(3) The map f satisfies the equation µ(df ) = 0.
In particular, the space of isotropic maps modulo V~ agrees with the
zeroes of µ in F0 via the bijection d : M /V~ → F0 .
2.11. Hamiltonian action. We now show that µ is indeed a moment
map:
Theorem 2.11.1. The action of T on F is Hamiltonian with moment
map µ. More precisely, µ is T-invariant and
Dhhµ, ζii = −ιXζ Ω
for every ζ ∈ t.
Proof. The invariance of µ is clear, by definition. The proof of the
theorem starts with a computation:
Lemma 2.11.2. For every F, Ḟ ∈ F , we have
Dµ|F · Ḟ = −g(RF, Ḟ ).
Proof. By bilinearity, 2Dµ|F · F = −g(RF, Ḟ ) − g(F, R Ḟ ) and he
lemma follows from the fact that R is pointwise g-selfadjoint.
For ζ ∈ t, we have Xζ (F ) = iζF by Formula (2.8) and it follows that
Ω(Xζ (F ), Ḟ ) = Ω(iζF, Ḟ )
= hhJ iζF, Ḟ ii
= hhζRF, Ḟ ii
Z
= g(ζRF, Ḟ )ωΣ
Σ
Z
= − ζDµ|F · Ḟ ωΣ
Σ
= −hhDµ|F · Ḟ , ζii,
which proves the theorem.
Proof of Theorem A and Corollary B. The restatement of the construc-
tions carried out at §2, together with Theorem 2.11.1 and Lemma 2.10.1
prove Theorem A. Corollary B is a restatement of Corollary 2.10.4.
2.12. Stability and isotropic maps. The Kempf-Ness theory, for Kähler
moment map geometry, relates the symplectic reduction with geomet-
ric invariant theory. This point of view seems appealing in our case,
where F is acted on by the complex gauge group TC and the µ is a
moment map for T. The question of existence of a zero of the mo-
ment map in a TC -orbit is rather trivial: for simplicity, we define the
TC -invariant subspace of generic differentials Fgen as the subspace of
nowhere vanishing differential forms. Then we have the following result
Lemma 2.12.1. For every F ∈ Fgen , the following properties are equiv-
alent:
(1) F + ∈ Fgen and F − ∈ Fgen .
(2) There exists λ ∈ TC such that µ(λ · F ) = 0.
In particular, the orbit of F ∈ F ± \ 0 does not contain any zero of the
moment map.
Proof. For λ : Σ → R \ 0 we have
1
µ(λ · F ) = − (λ−2 |F + |2 − λ2 |F − |2 )
2
and the lemma is obvious.
However, we are mostly interested in the zeroes of the moment map in
F0 , as they are differentials of isotropic maps by Corollary B. It would
be interesting to understand how the space of isotropic maps in M is
related to some notion of geometric stability on F . Unfortunately, the
image F0 = d(M ) is not T-invariant and it is not clear how to obtain
an analogue of Kempf-Ness theory from this point.
2.13. Energy of the moment map. We consider the energy of the mo-
ment map µ, given by the functional
φ:F R
(2.12)
F φ(F ) = 12 kµ(F )k2L2
Obviously, φ is non negative and
φ−1 (0) = µ−1 (0)
which is to say that the vanishing locus is the space of isotropic differ-
ential forms in F . By Corollary B, the vanishing locus of φ : F0 → R
is identified to the subspace of isotropic maps in M modulo the action
of V~ by translations via the correspondence d : M /V~ → F0 .
We prove various formulas about the differential and the gradient of
the functional φ on F :
Proposition 2.13.1. For every F, Ḟ ∈ F , we have
Dφ|F · Ḟ = −hhµ(F ), g(RF, Ḟ )ii,
Dφ|F · F = 4φ(F ),
and the gradient of the functional φ : F → R is given by the formula,
∇φ(F ) = −µ(F )RF.
or
∇φ = −J Z, (2.13)
where Z is the vector field on F defined by
Z(F ) = Xµ(F ) (F ).
Proof. By definition of φ(F ) = 12 kµ(F )k2 hence

Dφ|F · Ḟ = hhµ(F ), Dµ|F · Ḟ ii.
By Lemma 2.11.2, we have
Dφ|F · Ḟ = −hhµ(F ), g(RF, Ḟ )ii.
In particular, for Ḟ = F , we have Dφ|F · F = −hhµ(F ), g(RF, F )ii =
2hhµ(F ), µ(F )ii = 2kµ(F )k2 = 4φ(F ).
Now
Dφ|F · Ḟ = hhµ(F ), g(RF, Ḟ )ii
Z
= µ(F )g(RF, Ḟ )ωΣ
Σ
Z
= g(µ(F )RF, Ḟ )ωΣ
Σ
= hhµ(F )RF, Ḟ ii
and we deduce that
∇φ(F ) = −µ(F )RF.
In particular J ∇φ(F ) = −µ(F )J RF = µ(F )iF = Xµ(F ) (F ) and
we conclude that
∇φ = −J Z
where Z is the vector field on F defined by Z(F ) = Xµ(F ) (F ).
Corollary 2.13.2. The following properties are equivalent for F ∈ F0
(1) F is a zero of φ : F0 → R.
(2) F is a zero of µ in F0 .
(3) F is isotropic.
(4) F is a critical point of φ : F0 → R.
Proof. We know that (2) ⇔ (3) by Corollary 2.10.4 and we prove that
(1) ⇒ (2) ⇒ (4) ⇒ (1). If F is a zero of φ : F0 → R, it is a zero of µ,
by definition of φ. If µ(F ) = 0, then Dφ|F = 0 by Proposition 2.13.1,
hence F is a critical point of φ : F → R. This implies that F is also a
critical point of the restricted functional φ : F0 → R. If F is a critical
point of the restriction, then Dφ|F · F = 0. By Proposition 2.13.1, we
deduce that 0 = Dφ|F · F = 4φ(F ) and we conclude that φ(F ) = 0.
3. The modified moment map flow

The subspace of exact differentials F0 ⊂ F is not stable under the
T-action. Hence, the gradient ∇φ is generally not tangent to F0 . The
restriction of the functional φ : F0 → R admits a gradient vector field
∇◦ φ related to ∇φ by the formula
∇o φ(F ) = Π(∇φ(F )).
We define the modified moment map flow by the evolution equation

∂F
= −∇o φ(F ) (3.1)
∂t
along F0 and, more generally, along the Hölder completion F0k,ν .
3.1. Basic properties of the flow. An interesting feature of the flow is

the following decay property:
Theorem 3.1.1. Let Ft be a solution of the modified moment map flow,
for t in some interval of I. Then
∂
kFt k2L2 = −8φ(Ft ). (3.2)
∂t
In particular t 7→ kFt kL2 is non increasing along I.
Proof. We have
1∂ ∂Ft
kFt k2L2 = hh , Ft ii
2 ∂t ∂t
= −hh∇o φ(Ft ), Ft ii
= −hh∇φ(Ft ), Ft ii
= −4φ(Ft ).

Another important essential property is that the stationary points
of the modified moment map flow are exactly the zeroes of φ.
Proposition 3.1.2. The fixed point set of the modified moment map flow
agrees with the vanishing set of φ : F0 → R.
Proof. This is an immediate consequence of Corollary 2.13.2.
The modified moment map flow has the short time existence property
according to the following theorem:
Theorem 3.1.3. For every k ≥ 0, ν ∈ (0, 1) and F ∈ F0k,ν , there
exists ε > 0 and a unique solution of the modified moment map flow
Ft ∈ F0k,ν , defined for t ∈ [−ε, ε] such that F0 = F .
Proof. The map F 7→ ∇φ(F ) is polynomial of order 3 in the coefficients
of F by Proposition 2.13.1. In particular, it is locally Lipschitz with
respect to the C k,ν -norm. By Proposition 2.5.1, the Hodge projector
Π : F k,ν → F0k,ν is continuous for the C k,ν -norm. Therefore the maps
F 7→ ∇◦ φ(F ) = Π∇φ(F ) is locally Lipschitz on F0k,ν . The Cauchy-
Lipschitz theorem applies and the theorem is proved.
Proof of Theorem C. The result is just a restatement of Theorem 3.1.3,
Proposition 3.1.2 and Theorem 3.1.1.
3.2. Energy as a Morse-Bott function. By Proposition 2.13.1, we have

Dφ|F · Ḟ1 = −hhµ(F ), g(RF, Ḟ1)ii
for every F, Ḟ1 ∈ F and second variation of φ with respect to some
Ḟ2 ∈ F is given by
D 2 φ|F · (Ḟ1 , Ḟ2 ) = hhDµ|F · Ḟ1 , Dµ|F · Ḟ2 ii − hhµ(F ), g(R Ḟ2, Ḟ1 )ii.
In particular, if µ(F ) = 0, we have
D 2 φ|F · (Ḟ1 , Ḟ2 ) = hhDµ|F · Ḟ1 , Dµ|F · Ḟ2 ii (3.3)
which proves the following lemma:
Lemma 3.2.1. Let F ∈ F be an isotropic differential. Then, D 2 φ|F
is a non negative bilinear form. Furthermore, the kernel of D 2 φ|F is
given by the kernel of the linear map ker Dφ|F .
According to the above lemma, the functional φ : F0 → R seems
to behave like a Morse-Bott function. The rest of this section is de-
voted to proving that this is indeed the case, in a neighborhood of a
monomorphism.
Let s0 be the 0-section of the bundle T ∗ Σ → Σ. The map
Φ : Diff k,ν (Σ) × Ω1k,ν (Σ) → C k,ν (Σ, T ∗ Σ)
defined by Φ(ϕ, s) = s ◦ ϕ for some k ≥ 1 is smooth. Furthermore Φ
defines a local diffeomorphism between a neighborhood of (id, s0 ) and
a neighborhood of s0 in C k,ν (Σ, T ∗ Σ).
It is a well known fact that T ∗ Σ is endowed with a canonical symplec-
tic form ωT ∗ = dΛ, where Λ is the Liouville form on T ∗ Σ. Furthermore,
a map h : Σ → T ∗ Σ, sufficiently close to s0 in C 1 -norm is isotropic if,
and only if, h = Φ(φ, s) for some closed differential 1-form s [14].
It follows that the subspace of isotropic maps is a submanifold of
k,ν
C (Σ, T ∗ Σ) in a neighborhood of s0 . Furthermore, the subspace
of isotropic maps is locally diffeomorphic via Ψ to the submanifold
Diff k,ν (Σ) × Ω1,c 1,c
k,ν (Σ) in a neighborhood of (id, s0 ), where Ωk,ν (Σ) is the
subspace of closed forms.
We now specialize to the particular case where V has real dimen-
sion 4. Let f : Σ # V be an isotropic immersion. Then, the image
M = f (Σ) is known an immersed Lagrangian surface of V . By the
Lagrangian neighborhood theorem [14], there exists a neighborhood U
of the zero section s0 in T ∗ Σ and a neighborhood V of M in V , to-
gether with a smooth immersion θ : U → V , such that θ ◦ s0 = f and
θ ∗ ωV = ωT ∗ .
Every map f˜ : Σ → V , sufficiently close to f C 1 -norm, is given
by a map h : Σ → T ∗ Σ sufficiently close to s0 , such that f˜ = θ ◦ h.
Furthemore, f˜ is isotropic if, and only if, h is isotropic since θ preserves
the symplectic forms. We put Ψ(ϕ, s) = θ ◦ s ◦ ϕ and deduce the
following result:
Theorem 3.2.2. We assume that dimR V = 4 and let f : Σ # V be

an isotropic immersion with Hölder regularity C k,ν , for some k ≥ 1.
There exists smooth map
Ψ : Diff k,ν (Σ) × Ω1k,ν (Σ) → C k,ν (Σ, V )
defined in a neighborhood of Diff k,ν (Σ) × {s0 }, such that
(1) Ψ(id, s0 ) = f
(2) Ψ is Diff k,ν (Σ)-equivariant
(3) Ψ is a local diffeomorphism in a neighborhood of (id, s0 )
(4) The map Ψ restrict as a local diffeomorphism at (id, s0 ) between
Diff k,ν (Σ) × Ω1,c
k,ν (Σ) the isotropic deformations of f .
In particular, the subspace of isotropic deformations in a neighborhood
is a submanifold of C k,ν (Σ) in a sufficiently small open neighborhood
of f .
For k ≥ 1, the differential d induces a diffeomorphism
M k,α → F0k−1,α × V
given by f 7→ (df, f (x0 )), where x0 ∈ Σ is some fixed marked point. The
above diffeormorphism restrics to a diffeomorphism between isotropic
maps and istropic differentials and we deduce the following corollary:
Corollary 3.2.3. Assuming dimR V = 4 and k ≥ 0, let F ∈ F0k,ν be
an isotropic monomorphism, for some k ≥ 0. Then the subspace of
isotropic differentials is a submanifold of F0k,ν in a neighborhood of F .
Furthemore, the Hessian of φ is positive in transverse direction to the
submanifold of isotropic differential near F .
4. Polyhedral isotropic surfaces

In this section, we adapt all the smooth constructions of §2 and §3
to the polyhedral setting.
4.1. Definition of polyhedral surfaces. A triangulation of a surface Σ
is a triple T = (Σ, K , ℓ), where:
(1) Σ is a topological surface,
(2) K is a locally finite simplicial complex, contained in some am-
bient affine space A and
(3) ℓ : |K | → Σ is a homeomorphism, where |K | is the topological
space associated to the simplicial complex.
Given a triangulation T , every simplex σ ∈ K defines a subset
ℓ(σ) ⊂ Σ, homeomorphic to σ. It is often convenient to think of σ as a
domain in Σ, using the homeomorphism ℓ and we will take the liberty
to drop the reference to ℓ in our notations.
Every simplex σ of an affine space A spans an affine subspace of
A, with underlying vector space denoted ~σ , also called the tangent
direction of the simplex σ.
Let gΣ = (gσ )σ∈K , be a family, where gσ is a Euclidean inner product

on ~σ . Suppose that for every σ1 , σ2 ∈ K with σ2 ⊂ σ1 , the metric gσ2
agrees with the restriction of gσ1 to ~σ2 . In such a case, we say that gΣ
is a polyhedral metric on Σ.
Definition 4.1.1. A topological surface Σ endowed with a triangulation
T = (Σ, K , ℓ) and a polyhedral metric gΣ is called a polyhedral sur-
face.
The polyhedral metric gΣ provides a flat Riemannian metric gσ on
each simplex σ ∈ K . As in the smooth case discussed at §2.3, the
metric gσ induces a fiberwise Euclidean inner product, denoted gσ as
well, on all the tensor bundles over the simplex σ. Together with gV ,
we deduce a Euclidean fiberwise inner product for the bundles of V~ -
valued forms on σ, also denoted gσ . For simplicity of notation, these
inner products are sometimes denoted h·, ·i and the corresponding norm
is denoted | · |.
4.2. Whitney cohomology. The theory of Whitney forms and Whitney

cohomology [21] is a generalisation, in the piecewise linear setting, of
smooth differential forms and de Rham cohomology. The spaces of
smooth differential forms Ωn (σ) on a simplex σ ∈ K are compatible
with pullbacks in the following sense: for every σ1 , σ2 ∈ K with σ2 ⊂ σ1
and β1 ∈ Ωn (σ1 ), the pullback β2 of β1 on σ2 is a smooth differential
form as well.
We denote by Ωn (Σ, T ) the space of families of differential form
β = (βσ )σ∈K , where each βσ ∈ Ωn (Σ). Given β = (βσ ) ∈ Ωn (Σ, T ),
suppose that for every σ1 , σ2 ∈ K , with σ2 ⊂ σ1 , the pull back of βσ1
agrees with βσ2 . Then we say that β satifies the Whitney condition,
or that β is a Whitney form. The space of Whitney forms is denoted
Ωnw (Σ, T ).
There is a natural exterior derivative d : Ωn (Σ, T ) → Ωn+1 (Σ, T )
given by dβ = (dβσ )σ∈K . Furthermore, the spaces of Whitney forms
are preserved by d and we obtain the Whitney complex
d : Ωnw (Σ, T ) → Ωn+1
w (Σ, T )
which defines the Whitney cohomology denoted Hwn (Σ, R, T ). The

constructions readily extend to the case of V~ -valued differential forms
and we obtain the Whitney cohomology spaces Hwn (Σ, V~ , T ). Similarly
to the de Rham cohomology, the Whitney cohomology agrees with the
usual cohomology spaces H n (Σ, R) ≃ Hwn (Σ, R, T ) and H n (Σ, V~ ) ≃
Hwn (Σ, V~ , T ). In particular, the Whitney cohomology does not depend
on the choice of triangulation T .
4.3. Orientation and Kähler structure. An orientation of the simpli-

cial complex K of a polyhedral surface Σ is also called an orientation
of Σ. An oriented polyhedral surface carries a canonical area form

ωΣ = (ωσ )σ∈K ∈ Ω2w (Σ, T ) characterized by
(1) ωσ is an exterior 2-form on ~σ .
(2) For every σ ∈ K2 , the volume form ωσ is compatible with the
orientation of the facet and |ωσ | = 1, with respect to gσ .
(3) If σ ∈ K is an edge or a vertex then ωσ = 0.
The combination of ωσ and gσ defines a family of almost complex
structures JΣ = (Jσ )σ∈K2 on the facets of K , where Jσ : ~σ → ~σ is an
almost complex structure on ~σ , with the property that ωσ = gσ (Jσ ·, ·).
In conclusion, an oriented polyhedral surface Σ is endowed with the
structures (T , gΣ , JΣ , ωΣ ), referred to as a polyhedral Kähler structure,
a notion introduced by Dmitri Panov in [16], where he investigates the
4 dimensional case.
Remark 4.3.1. A polyhedral Kähler surface Σ as above admits a smooth
structure [16, 19, 20] such that
(1) gΣ and ωΣ are smooth away from the vertices of the triangula-
tion.
(2) The metric may have conical singularities at the vertices.
(3) The almost complex structure JΣ is smooth on Σ.
In presence of conical singularities, the map ℓ : σ → Σ is not smooth at
the vertices. In particular, the smooth differential forms on Σ do not
necessarily induce smooth families of Whitney forms in Ωnw (Σ, T ).
4.4. Polyhedral maps and differentials. We continue our constructions
and analogies with the smooth setting, when Σ is a closed oriented
polyhedral surface.
A map f : Σ → V , or more generally a map to some affine space, is
called polyhedral with respect to T if:
(1) the map f is continuous and
(2) the restriction of f to every simplex of the triangulation T is
an affine map.
The space of polyhedral maps is denoted
M (T ) = {f : Σ → V, f is T -polyhedral}.
The restriction of a polyhedral map f to every simplex σ ∈ T is affine,
hence differentiable. However, f is generally not differentiable at every
point of Σ. Nevertheless, we can define the differential df as a family
df = (df |σ )σ∈K .
By construction, df is a Whitney form df ∈ Ω1w (Σ, V~ , T ). The moduli

space M (T ) is an affine space with underlying vector space contained
in Ω0w (Σ, T , V~ ) and it follows that df is an exact Whitney form.
These observations motivate the introduction of a space of constant

V -valued differentials, closely related to Ω1 (Σ, V~ , T ):
~
F (T ) = {(Fσ )σ∈K2 , Fσ ∈ ~σ ∗ ⊗ V~ },
where ~σ ∗ is the dual of the tangent direction ~σ . An element of ~σ ∗ ⊗ V~
is a linear map Fσ : ~σ → V~ and this map can be regarded as a constant
V~ -valued differential 1-form on σ.
However, there are no compatibility conditions a priori along the
edges of the triangulation. We introduce the subspace of Whitney forms
which is a slightly different condition since σ ∈ K2 : an element F =
(Fσ )σ∈K2 ∈ F (T ) satisfies the Whitney condition if for every σ1 , σ2 ∈
K2 with a common edge σ3 , the pullbacks of Fσ1 and Fσ2 agree along
σ3 . In this case, F is called a Whitney form and we define:
Fw (T ) = {F ∈ F (T ), F is a Whitney form}.
Thanks to the Whitney condition, a Whitney form F ∈ Fw (T ) de-
fines via the pullback an extended family (Fσ ) for σ ∈ K which is
understood as an element of Ω1w (Σ, V~ , T ). Hence, we have a canonical
embedding
Fw (T ) ֒→ Ω1w (Σ, V~ , T ).
The elements F ∈ Fw (T ) are families of constant differential forms
along the vertices of T , hence they are closed. Thus, every F defines
a Whitney cohomology class denoted [F ] ∈ Hw1 (Σ, V~ ). We define the
subspace of exact forms in Fw (T ) by
F0 (F ) = {F ∈ Fw (T ), [F ] = 0}.
By construction, the differential d defines a map
d : M (T ) → F0 (T ).
Furthermore the map d induced an homeomorphism
d : M (T )/V~ → F0 (T ).
where V~ acts by translations on M (T ).
4.5. Moduli space structure in the polyhedral setting. The space F (T )
carries a Kähler structure defined similarly to the smooth setting. For
F = (Fσ ) and H = (Hσ ) ∈ F (T ), we put
XZ
G(F, H) = hFσ , Hσ iωσ .
σ∈K2 σ
The integrands of the above integrals are constant on each simplex.

Hence each term of the above integral is equal to hFσ , Hσ iarea(σ),
where area(σ) is the area of σ with respect to the Euclidean metric
gσ . As in the smooth case, we will use the notation hhF, Hii = G(F, H)
and kF kL2 for the norm deduced from G.
An almost complex structure J on F (T ) is given by

(J F )σ = −Fσ ◦ Jσ .
By construction J is compatible with G and we obtain a Kähler form
Ω = G(J ·, ·).
Finally, we have defined a flat Kähler structure
(F (T ), G, J , Ω)
on the moduli space.
As in the smooth case, the almost complex structure i acts by mul-
tiplication on V~ -valued forms and we obtain a linear involution R on
F (T ), defined by
RF = iJ F.
The involution gives an orthogonal splitting into eigenspaces
F (T ) = F + (T ) ⊕ F − (T )
and we introduce the G-orthogonal projection onto the subspace of
exact differentials
Π : F (T ) → F0 (T ).
4.6. Polyhedral symplectic density. We denote denote by tC (T ) the

space of maps ζ : K2 → C. Equivalently, an element ζ can be un-
derstood as a family of constant maps (ζσ )σ∈K2 on each facet σ of the
triangulation. We also denote by t(T ) the space of real valued maps
ζ : K2 → R and we define
µ : F (T ) → t(T )
by the formula
1
µ(F )(σ) = − h(RF )σ , Fσ i.
2
We can make sense of the pullback F ∗ ωV as a family of differential
2-forms along each facet defined by
(Fσ∗ ωV )(η1 , η2 ) = ωV (Fσ · η1 , Fσ · η2 )
for every η1 , η2 ∈ ~σ , and σ ∈ K2 . As in the smooth case, the map µ is
a symplectic density, in the sense of the following lemma:
Lemma 4.6.1. For every F ∈ F (T ), we have the identity
Fσ∗ ωV
µ(F )(σ) = − (4.1)
ωσ
for every facet σ ∈ K2 .
This motivates the following definition:
Definition 4.6.2. A differential F ∈ F (T ) is called isotropic if Fσ :

~σ → V is isotropic for every σ ∈ K2 . A polyhedral map f ∈ M (T ) is
called isotropic if the pullback of ωV by f vanishes along every simplex
of the triangulation.
In particular, by Lemma (4.6.1), we have
Lemma 4.6.3. A differential F ∈ F (T ) is isotropic if, and only if,

µ(F ) = 0. A polyhedral map f ∈ M (T ) is isotropic if, and only if,
F = df is isotropic.
4.7. Hamiltonian gauge group action. We define the complex gauge

group TC (T ) as the space of non vanishing complex valued functions
λ : K2 → C∗ . Alternatively, λ can be thought of as a family of constant
functions (λσ ) along each facet of the triangulation. The group TC (T )
acts on F (T ) via the formula
λ · F = λ̄−1 F + + λF − , (4.2)
where F ± are the component of F given by the splitting F (T ) =

F + (T ) ⊕ F − (T ). The real subgroup T(T ) is the the space of func-
tions λ : K2 → S 1 and acts by complex multiplication on F (T ). The
exponential map
exp : tC (T ) → TC (T )
defined by
exp ζ = eiζ ,
shows that tC (T ) is the Lie algebra of TC (T ) and identifies the Lie
algebra of the subgroup t(T ) with t(T ). The infinitesimal action of
tC (T ) on F (T ) is given by
Xζ (F ) = iζ̄F + + iζF − .
As in the smooth case, the construction has the following nice proper-
ties:
Proposition 4.7.1. The TC (T )-action on F (T ) perserves the almost

complex structure J . Furthermore, the TC (T )-action is the J -complexification
of the T(T )-action. The T-action preserves the metric G and the sym-
plectic form Ω as well. Furthermore the T(T )-action is Hamiltonian
with moment map µ : F (T ) → t(T ) given by Formula (4.1). More
precisely, µ is T(T )-invariant and for every ζ ∈ t(T ), we have
Dhhµ, ζii = −ιXζ Ω.
Proof. The proof is formally identical to the smooth setting.

4.8. Polyhedral modified moment map flow. By analogy with the smooth
setting, we consider the energy of the moment map
φ : F (T ) → R
defined by
1
φ(F ) = kµ(F )k2L2 .
2
By contruction, φ is non negative and its vanishing set is the subspace
of isotropic differentials in F (T ). We denote by ∇φ the gradient
of φ : F → R with respect to the metric G. The gradient of the
restricted functional φ : F0 (T ) → R is denoted ∇◦ φ and satisfies
∇◦ φ(F ) = Π(∇φ(F )). Then we define the polyhedral modified moment
map flow by the evolution equation along F0 (T ):
∂F
= −∇◦ φ(F ). (4.3)
∂t
As in the smooth case, the formal identities are the same and we can
prove:
Proposition 4.8.1. The critical points of the restricted functional φ :
F0 (T ) → R, in other words the fixed points of the modified moment
map flow, agree with its vanishing locus.
For every F ∈ F0 (T ), there exists a polyhedral map f ∈ M (T )
such that F = df , unique up to the action of V~ with the property that
f is isotropic if, and only if, φ(F ) = 0.
Remark 4.8.2. The Second statement of the proposition implies that
the modified moment map flow defines a flow for polyhedral maps as
well: we fix a point x0 ∈ Σ and a point v0 ∈ V . For every F ∈ F0 (F ),
we define the integral f = χ(F ) ∈ M (T ) as the unique map such that
df = F and f (x0 ) = v0 . Any solution Ft of the modified moment map
flow defines a family ft = χ(Ft ) solution of the evolution equation
∂f
= −χ ◦ ∇◦ φ(df ) (4.4)
∂t
and vice-versa. The fixed points of this flow are, by definition, the
isotropic polyhedral maps.
Proposition 4.8.3. A solution of the polyhedral modified moment map
flow Ft ∈ F0 (T ), defined for t in some interval satisfies
∂
kF k2L2 = −8φ(Ft ).
∂t
In particular, the L2 -norm is non increasing along the interval.
Proof. The ingredients of the proof are the same as in Theorem 3.1.1,
for the smooth setting. In the polyhedral context, we have the identity
Dφ|F · F = 4φ(F ) (4.5)
identical to the one in Proposition 2.13.1 and the rest of the argument
is identical.
Proposition 4.8.3 has the following strong consequence:
Corollary 4.8.4. For every F ∈ F0 (T ), there exists a unique solution

Ft defined for t ∈ [0, +∞) and such that F0 = F . Furthermore, kFt kL2
is bounded by kF kL2 .
Proof. The short time existence is immediate Proposition 4.8.3 and

classical ODE theory. The decay of the L2 -norm implies that a solution
of the flow cannot blow up in finite time, which implies the long time
existence.
4.9. A Duistermaat theorem. In fact, the polyhedral modified moment

map flow converges and provides a strong deformation retraction onto
the space of polyhedral isotropic maps.
Theorem 4.9.1. For every F ∈ F0 (T ), the solution Ft of the modified

moment map flow, defined for t ∈ [0, +∞) with F0 = F converges
toward a limit F∞ such that φ(F∞ ) = 0. Thus, we can define an
extended flow
Θ : [0, +∞] × F0 (T ) → F0 (T )
by Θ(t, F ) = Ft and Θ(+∞, F ) = F∞ . Furthermore, the map Θ is a
strong deformation retraction onto the vanishing locus of φ : F0 (T ) →
R.
Proof. The proof is essentially contained in the analogous result [18,

Theorem 7.8.4], where a downward gradient flow is also studied in a
finite dimensional situation. Although the space F0 (T ), the functional
φ : F0 (T ) → R and the Euclidean metric G are completely different
in [18], they share similar formal properties:
(1) The vector space is finite dimensional.
(2) The functional φ is a polynomial function.
(3) The functional φ is non negative.
(4) The critical points of φ agree with its vanishing set.
(5) The solutions of the downward gradient flow satisfy the decay
property of Proposition 4.8.3 .
Properties (1) and (5) imply the long time existence of the flow proved
at Corollary 4.8.4. Property (2) implies that φ is analytic. In particular
the Lojaziewicz inequality can be applied to φ, which is crucial to prove
the convergence of the flow lines. The interested reader may check that
only these properties are used in the proof of [18, Theorem 7.8.4] and
that they imply the congergence and continuity of the flow.
4.10. Regular points of the moduli space. We obtain much stronger

result for the behavior of the modified moment map flow in the polyhe-
dral setting, as proved by Theorem 4.9.1. On the contrary, the regular-
ity of the space of isotropic maps is much easier to study in the smooth
setting, at least in a neighborhood of an immersion, as showed at §3.2.
However, some partial results can be obtained in the polyhedral case,
as we are going to see in this section.
Definition 4.10.1. A differential form F ∈ F0 (T ) is called regular, if
the differential of the restricted moment map µ : F0 (T ) → t(T ) has
constant rank in a neighborhood of F .
By definition, the vanishing locus of φ : F0 (T ) → R is a submanifold
of F0 (T ) in a neighborhood of a regular point F . Furthermore, the
following theorem shows that φ behaves like a Morse-Bott function on
such a neighborhood:
Theorem 4.10.2. Let F ∈ F0 (T ) be a regular point of the vanishing
locus of φ. Then, there exists an open neighborhood U ⊂ F0 (T ) of F ,
such that:
(1) the vanishing locus of φ : U → R is a submanifold of U;
(2) the Hessian of φ is positive definite in direction transverse to
the vanishing locus;
(3) U is invariant under the modified moment map flow Θ, which
has an exponential convergence rate.
Proof. The Hessian of φ is calculated in the smooth setting at §3.2
and is given a point F such φ(F ) = 0 by Formula (3.3). The same
calculations show that the formula holds in the polyhedral setting as
well and obtain
D 2 φ|F · (Ḟ1 , Ḟ2 ) = hhDµ|F · Ḟ1 , Dµ|F · Ḟ2 ii
a any point F of the vanishing locus of φ : F0 (T ) → R. If F is a regu-
lar, the vanishing locus of φ : F0 (T ) → R is a submanifold of F0 (T )
in a neighborhood of F and the above formula shows that the Hessian
of φ is definite positive in transverse directions to the submanifold. In
conclusion, φ behaves like a Morse-Bott function in a neighborhood
of F and its vanishing set is a stable critical component. The state-
ments of the theorem are classical facts of ODE theory under these
assumptions.
Constructing regular points of the moduli space is not a simple prob-
lem. We are going to show that Theorem 1.1.1 provides a construction
of regular points. In [11], a quotient 2-torus Σ = R2 /Γ is considered
together with a smooth isotropic immersion f : Σ # R. Refinements of
stepsize O(N −1 ) of the lattice Γ provide a family of quadrangulations
QN of Σ, with N 2 facets. The quadrangulations can be completed into
triangulations TN by replacing each quadrilateral of QN with a pyra-

mid. The main construction of [11] provides a sequence of polyhedral
isotropic maps fN ∈ M (TN ), for every N sufficiently large, such that
fN approximates f in C 1 -norm by [17]. The construction of fN relies
on an effective version of the fixed point principle applied to the family
of symplectic densities
µN ◦ d : M (TN ) → t(TN ),
where µN : F0 (TN ) → t(TN ) is the moment map. The fact that the
cohomology class of ωV vanishes implies that µN ◦ d takes values in
the subspace of ζ ∈ t(TN ) orthogonal to constant functions. Hence
µN ◦ d can never be a submersion, but it may have maximal rank with
corank one. A careful examination of the proof in [11] allows to make
the following observation:
Lemma 4.10.3. For every N sufficiently large, the differential of the
map µN ◦ d : M (TN ) → t(TN ) has maximal rank at fN , where fN is
the Jauberteau-Rollin-Tapie approximation of f . In particular, FN =
dfN ∈ F0 (TN ) is a regular point of the vanishing set of φ : F0 (TN ) →
R.
We can now complete the proofs of the main theorems given in the
introduction.
Proof of Theorem G. The result is an immediate consequence of Lemma
4.10.3.
Proof of Theorem E. The result is a combination of Proposition 4.8.1,
Theorem 4.9.1 and Theorem 4.10.2. The statement (3) will be proved
at the end of §4.11.
Proof of Corollary F. The differential d : M (T ) → F0 (T ) admits a
unique right inverse
χ : F0 (T ) → M (T )
defined by the condition that d ◦ χ = id and f = χ(F ) satisfies f (x0 ) =
v0 for some marked points x0 ∈ Σ and v0 ∈ V . We consider the
continuous map
Θ̂(t, f ) = χ ◦ Θ(t, df ) + (f (x0 ) − v0 ).
By construction
Θ̂ : [0, +∞] × M (T ) → M (T )
defines a strong deformation retraction of M (T ) onto the subspace of
isotropic polyhedral maps. It is easy to see that Θ̂ is the flow of the
evolution equation on M (T ) defined at Remark 4.8.2. In particular
Θt ◦ d = d ◦ Θ̂t , which proves the corollary.
4.11. Topology of the moduli space. The homotopy of the space of

smooth isotropic immersions f : Σ # V can be reduced to a mat-
ter of algebraic topology, thanks to the Gromov-Lees theorem [9, 13]:
every isotropic immersion f defines an exact differential form F =
df ∈ Ω1 (Σ, V~ ) which is by definition an isotropic monomorphism. The
Gromov-Lees theorem establishes an h-principle, which shows that the
differential d : M /V~ → F0 induces a homotopy equivalence from the
space of isotropic immersions to the space of isotropic monomorphisms.
Computing the homotopy of the latter space turns out to be a basic
problem of algebraic topology, related to the Grassmanian of isotropic
planes in V~ .
Unfortunately, the Gromov-Lees theory does not extend immediately
to the polyhedral setting. Thus, most topological properties for the
space of polyhedral isotropic immersions are open questions although
the partial results of [11, 17, 6] lead us to state Conjecture H.
The map d : M (T )/V~ → F0 (T ) is a homeomorphism, that identi-
fies the subspace of polyhedral isotropic maps modulo V~ with the space
of exact isotropic differentials. The latter space is a cone in F0 (T ),
which implies that it is contractible. Therefore, the space of polyhedral
isotropic maps is homotopically trivial. The same property holds in the
smooth setting, which is why the problem of immersions is considered
instead. There are several issues here to formulate analogous questions
in the polyhedral setting:
(1) A polyhedral map f ∈ M (T ) which is a topological immer-
sion define a monomorphism F = df ∈ F0 (T ). However the
converse is generally not true and some local injectivity condi-
tion along the skeleton of the triangulation should be added to
obtain a correspondence.
(2) The space of monomorphisms in F0 (T ) does not seem to be
preserved by the modified moment map flow Θ. From this point,
the prospect of using the flow to investigate the homotopy type
of the space of polyhedral isotropic immersions seems rather
low.
However the space of non constant polyhedral isotropic maps may al-
ready contain some non trivial topology. Furthemore, this subspace
modulo V~ is identified to F0 (T ) \ 0 via d. Finally, F0 (T ) \ 0 is in-
variant under the gauge group action and invariant under the modified
moment map flow, for finite time.
Question 4.11.1. What is the homotopy type of the space of non con-
stant polyhedral isotropic maps f ∈ M (T ) ?
The space of isotropic exact differentials in F0 (T ) \ 0 is invariant by

scaling. In particular, it is homotopically equivalent to the vanishing
set of the restriction φ : S(T ) → R, where S(T ) is the unit sphere in

F0 (T ).
The idea to tackle Question 4.11.1 is to interpret φ : S(T ) → R as
a Morse-Bott function to obtain some dynamical interaction between
its critical components. By definition the critical set of φ : S(T ) → R
contains its vanishing set, but there may be other critical points, called
the solitons. More explicitly, a soliton F ∈ F0 (T ) is a solution of the
equation ∇◦ φ(F ) = κF for some κ ∈ R. The constant κ is determined
by Formula 4.5 and we obtain the soliton equation:
kF k2L2 ∇◦ φ(F ) = 4φ(F ). (4.6)
Solitons are by definition the fixed points of the downward gradient
flow of the restricted functional φ : S(T ) → R defined for F ∈ S(T )
by
∂F
= 4φ(F ) − ∇◦ φ(F ) (4.7)
∂t
and called the renormalized polyhedral flow.
Many questions are open at this stage for the renormalized flow and
the solitons:
(1) Is φ : S(T ) → R a Morse-Bott type function, in some sense,
perhaps up to the choice a refinement of the triangulation and
a generic polyhedral metric ?
(2) What are the intrinsic geometrical properties of polyhedral maps
f ∈ M (T ) such that F = df is a soliton ?
(3) Does the functional φ : S(T ) → R define a Morse-Bott coho-
mology theory ?
We can prove that solitons never reduce to the vanishing locus of φ.
Lemma 4.11.2. There exists solitons G ∈ F0 (T ) such that φ(G) 6= 0.
Proof. Clearly, φ : S(T ) → R is not the zero map, as it is easy to
construct polyhedral maps f ∈ M (T ) which are not isotropic. In par-
ticular φ(df ) > 0 in this case. We conclude that φ admits a maximum
at some point G ∈ S(T ) with φ(G) > 0. If follows that G is a crit-
ical point of φ : S(T ) → R, which is to say a soliton. Furthermore,
φ(G) > 0 implies that G is not isotropic.
Remark 4.11.3. We introduce a family Ft ∈ F0 (T ) given by Ft =
r(t)G, where G ∈ F0 (T ) is a soliton such that φ(G) > 0 and
1
r(t) = p
8(t − t0 )φ(G)
is defined on the interval (t0 , +∞), for some choice of t0 ∈ R. It is
readily checked that Ft is a solution of the polyhedral modified moment
map flow. We notice that Ft defines a non trivial solution of flow, with
the property that lim Ft = 0.
t→+∞
Proof of Theorem E, item (3). The existence of non isotropic solitons

in S(T ) provides non trivial solutions of the polyhedral modified mo-
ment map flow as discussed above. This completes the proof of the
theorem.
4.12. Numerical flow. The polyhedral modified moment map flow is
an ordinary differential evolution equation. A numerical version of the
flow can be implemented on a computer. In this section we outline
all the ingredients for such a computer program. The code is being
developed currently and is going to be released very soon [10].
For simplicity, we specialize to the case where Σ is a quotient 2-torus.
Let Γ be the lattice
Γ = γ1 Z ⊕ γ2 Z ⊂ C
i π3
where γ1 = 1 γ2 = e . The surface Σ, defined by
Σ = C/Γ,
is endowed with the flat Kähler structure (Σ, gΣ , JΣ , ωΣ ) deduced from
the canonical flat Kähler structure of C. The rhombus in C with ver-
tices 0, γ1 , γ1 + γ2 and γ2 is a fundamental domain for the action of Γ.
This rhombus
√
is composed of two equilateral triangles and each triangle
3
has area 4 . Using the Γ-action, we obtain the familiar tiling of C by
equilateral triangles:
Figure 1. Triangulation of C by equilateral triangles
This tiling is a simplicial decomposition of C and can be regarded

as a triangulation T ′ of C. Let N be a positive integer. We define the
lattice
ΓN = N −1 Γ,
understood as the set of vertices of the rescaled triangulation TN′ of T ′ ,
by a factor N −1 . By Definition, the triangulation TN′ is Γ-invariant.
Hence, TN′ defines a triangulation
√
of the quotient Σ, denoted TN , with
3
each triangle of area 4N 2 .
We consider the particular space V = C2 endowed with its canonical
structure of Hermitian affine space and the coordinates (x1 + ix2 , x3 +
ix4 ) ∈ V , where xi ∈ R. The symplectic form ωV is given in coordinates

by the formula
ωV = dx1 ∧ dx2 + dx3 ∧ dx4
and the almost complex structure i : V~ → V~ , given by the multiplica-
tion by i satisfies the identities
∂ ∂ ∂ ∂
i = and i = . (4.8)
∂x1 ∂x2 ∂x3 ∂x4
Every F ∈ F (TN ) is given by a collection F = (Fσ )σ∈K2 where
Fσ ∈ ~σ ∗ ⊗ V~ . We use the canoncial coordinates z = u1 + iu2 ∈ C on
the universal cover C of Σ. By definition we have
−du1 ◦ JΣ = du2 and − du2 ◦ JΣ = −du1 (4.9)
The dui ⊗ ∂x∂ j provide a gσ -orthonormal basis of ~σ ∗ ⊗ V~ for each facet
of the triangulation TN . Hence, Fσ admits a decomposition
X ∂
Fσ = Fσij dui ⊗ (4.10)
1≤i≤2
∂x j
1≤j≤4
where Fσij ∈ R. In particular, for F, H ∈ F (TN ), we have

√
X 3
hhF, Hii = Fσij Hσij
1≤i≤2
4N 2
1≤j≤4
σ
where the sum is taken over all the facets σ of TN .

The space F0 (TN ) is the subspace of exact V~ -valued Whitney forms
in F (TN ). We need to construct an orthonormal basis of this subspace
to compute the matrix of the projector Π : F (TN ) → F0 (TN ).
We consider the affine canonical base of V given by the points e0 = 0
and ej = e0 + ∂x∂ j for 1 ≤ j ≤ 4. For each vertex σ0 of TN and 1 ≤ j ≤ 4,
we define the polyhedral function f σ0 j : Σ → V given by

σ0 j ′ ej if σ0 = σ0′ ,
f (σ0 ) =
e0 otherwise.
for every vertex σ0′ of TN . The above function extends uniquely as a
polyhedral function f σ0 j ∈ M (TN ) and defines the exact differential
F σ0 j = df σ0 j ∈ F0 (TN ).
The family F σ0 j for σ0 ∈ K0 and 1 ≤ j ≤ 4 spans F0 (TN ), by def-
inition. The only relations between the F σj are the 4 relations given
by X
F σ0 j = 0
σ0
for 1 ≤ j ≤ 4, where the sum is taken over all the vertices of the
triangulation TN . We obtain the following lemma:
Lemma 4.12.1. For a fixed vertex σ0′ of the triangulation TN , The fam-
ily
F σ0 j ∈ F0 (TN ),
where 1 ≤ j ≤ 4 and σ0 belongs to the vertices of TN with σ0 6= σ0′ , is
a basis of F0 (TN ).
However, the family F σ0 j is not orthonormal and we have to use
the Gram-Schmidt algorithm. The differential F σ0 j can be computed
explicitly, working on the universal cover C of Σ, and assuming σ0 = 0
for simplicity. There are six triangles of TN′ with vertex σ0 = 0. We
can compute the differential of f σ0 j on each triangle. In particular on
the triangle σ1 with vertices 0, γ1 /N and γ2 /N we have

σ0 j N ∂
Fσ1 = −Ndu1 − √ du2 ⊗ . (4.11)
6 ∂xj
The differential on all the other triangles with vertex σ0 = 0 is obtained
by rotations of angle π3 . For every other triangle σ not in the star of
σ0 , we have Fσσ0 j = 0. In particular, on the triangle σ2 with vertices 0,
π
−γ1 /N and e2i 3 /N, we have the formula

σ0 j N ∂
Fσ2 = Ndu1 − √ du2 ⊗ . (4.12)
6 ∂xj
Lemma 4.12.2. The family of Whitney forms F σ0 j ∈ F0 (TN ) described
above satisfies the identities
′ ′
hhF σ0 j ,F σ0 j ii =
 7√3
 4 if i = i′ and σ0 = σ0′ ,
5
− √ if i = i′ and σ0 and σ0′ are joined by an edge,
 4 3
0 otherwise.
σ′ j ′
Proof. The third case is clear, since the families (Fσσ0 j ) and (Fσ 0 ) have
disjoint supports for j 6= j ′ or if σ0 and σ0′ do not belong to the same
facet of TN .
We assume now that, on the contrary, j = j ′ and σ0 and σ0′ belong
to the same facet. Then either σ0 = σ0′ or σ0 and σ0′ are the two ends
of an edge in TN .
In the first case, working on the universal cover and assuming σ0 = 0,
σ0 j 2 7N 2
Formula (4.11) √
shows that |Fσ 1
| = 6
. Using the fact that the area
3
of a facet is 4N 2 and that there are 6 facets in the star of σ0 , we find
√
σ0 j 2 7 3
kF kL2 = .
4
In the second case, working on the universal cover, we may assume that
σ′ j
σ0′ = γN1 . By Formulas (4.11) and (4.12), we deduce that hFσσ10 j , Fσ10 i =
2
− 5N6 . There is a second facet of TN in the common support of F σ0 j
′
and F σ0 j . By symmetry, the inner product is the√same on the second
facet. Since the area of each triangle is equal to 4N32 , we obtain
2
√
′ 5N 3 5
hhF σ0 j , F σ0 j ii = − ·2· 2
=− √ .
6 4N 4 3

The family F σ0 j provides a basis of F0 (TN ) thanks to Lemma 4.12.1
and the Gram-Schmidt method gives an algorithm to construct an or-
thonomal basis on F0 (TN ) from this data. Furthermore, the Gram-
Schmidt method is pretty straightforward regarding computer imple-
mentation thanks to the explicit formulas of Lemma 4.12.2.
Eventually, the matrix of the endomorphism
R : FN (FN ) → F (TN )
can be explicitely computed in the coordinates given by (4.10), using
Formula (4.9) and Formula (4.8) as RF = −iF ◦ JΣ , by definition.
Thus, we have an algorithm to compute µ(F ) = − 12 hRF, F i as well
as ∇φ(F ) = −µ(F )RF . We also have a matrix representation for Π
via Gram-Schmidt and we can compute ∇◦ φ(F ) = Π(∇(F )). It is now
easy to find approximate solutions of the polyhedral modified moment
map flow (4.3) via the Euler method. Similarly we can apply the Euler
approximation method to the polyhedral renormalized flow (4.7).
References
[1] M. F. Atiyah and R. Bott. The Yang-Mills equations over Riemann surfaces.
Philos. Trans. R. Soc. Lond., Ser. A, 308:523–615, 1983.
[2] A. Banyaga. The structure of classical diffeomorphism groups, volume 400 of
Math. Appl., Dordr. Dordrecht: Kluwer Academic Publishers, 1997.
[3] M. Bertelson and J. Distexhe. PL approximations of symplectic manifolds. J.
Symplectic Geom., 2024. to appear.
[4] S. K. Donaldson. Moment maps and diffeomorphisms. Asian J. Math., 3(1):1–
15, 1999. Sir Michael Atiyah: a great mathematician of the twentieth century.
[5] Y. Eliashberg and N. Mishachev. Introduction to the h-principle, volume 48
of Graduate Studies in Mathematics. American Mathematical Society, Provi-
dence, RI, 2002.
[6] S. Etourneau. Approximation C 1 d’immersions isotropes lisses par des immer-
sions isotropes PL. Nantes University, PhD thesis, 2023.
[7] B. Gratza. Piecewise linear approximations in symplectic geometry. ETH
Zurich, PhD thesis, 1998.
[8] M. Gromov. Pseudo holomorphic curves in symplectic manifolds. Invent.
Math., 82:307–347, 1985.
[9] M. Gromov. Partial differential relations, volume 9 of Ergeb. Math. Grenzgeb.,
3. Folge. Springer, Cham, 1986.
[10] F. Jauberteau and Y. Rollin. Numerical flow for polyhedral isotropic surfaces.
In preparation, 2024.
[11] F. Jauberteau, Y. Rollin, and S. Tapie. Discrete geometry and isotropic sur-
faces. Mém. Soc. Math. Fr., Nouv. Sér., 161:1–99, 2019.
[12] D. D. Joyce. Compact manifolds with special holonomy. Oxford Math. Monogr.
Oxford: Oxford University Press, 2000.
[13] J. A. Lees. On the classification of Lagrange immersions. Duke Math. J.,
43:217–224, 1976.
[14] D. McDuff and D. Salamon. Introduction to symplectic topology, volume 27 of
Oxf. Grad. Texts Math. Oxford: Oxford University Press, 3rd edition edition,
2016.
[15] D. Mumford, J. Fogarty, and F. Kirwan. Geometric invariant theory., vol-
ume 34 of Ergeb. Math. Grenzgeb. Berlin: Springer-Verlag, 3rd enl. ed. edition,
1994.
[16] D. Panov. Polyhedral Kähler manifolds. Geom. Topol., 13(4):2205–2252, 2009.
[17] Y. Rollin. Polyhedral approximation by Lagrangian and isotropic tori. J. Sym-
plectic Geom., 20(6):1349–1383, 2022.
[18] Y. Rollin. Symplectic maps and hyperKähler moment map geometry, 2024.
[19] W. P. Thurston. Shapes of polyhedra and triangulations of the sphere. In The
Epstein Birthday Schrift dedicated to David Epstein on the occasion of his
60th birthday, pages 511–549. Warwick: University of Warwick, Institute of
Mathematics, 1998.
[20] M. Troyanov. Les surfaces euclidiennes à singularités coniques. (Euclidean sur-
faces with cone singularities). Enseign. Math. (2), 32:79–94, 1986.
[21] H. Whitney. Geometric integration theory, volume 21 of Princeton Math. Ser.
Princeton University Press, Princeton, NJ, 1957.
François Jauberteau, Laboratoire Jean Leray, Université de Nantes

Email address: francois.jauberteau@univ-nantes.fr
Yann Rollin, Laboratoire Jean Leray, Université de Nantes

Email address: yann.rollin@univ-nantes.fr

2404 11347

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2404 11347

Uploaded by

Copyright:

Available Formats

ISOTROPIC SURFACES AND MOMENT MAP FLOW

FRANÇOIS JAUBERTEAU AND YANN ROLLIN

Abstract. We consider the moduli space of isotropic maps from

Kähler moment map geometry, on a space of differential forms on

2010 Mathematics Subject Classification. Primary 5299, 53D12; Secondary

Theorem 1.1.1 ([17]). Let f : Σ # V be a smooth isotropic immersion,

A Kähler moment map geometry inspired by Donaldson [4] was in-

The image of M by d is the space of exact 1-forms denoted F0 . We

We consider the energy functional of the moment map

The purpose of §4 is to extend all the above constructions to the

the functional φ restricted to F0 (T ) is called the polyhedral modified

Theorem E. For F ∈ F0 (T ), the polyhedral modified moment map

defined by Θ(t, F ) = Ft for t ∈ [0, +∞) and Θ(+∞, F ) = F∞ has the

The ﬂow Θ on F0 can be lifted as a ﬂow Θ̂ on M (T ) and we have

Corollary F. There exists a unique map

The existence of regular points (cf. Deﬁnition 4.10.1) of the vanishing

Theorem G. In the case where Σ is a 2-torus and f : Σ # V is

2.1. Target space 9

2. Kähler moment map

2.1. Target space. Let V be an affine space and V~ , its underlying

hV (v1 , v2 ) = gV (v1 , v2 ) + iωV (v1 , v2 ),

into the real and imaginary parts of hV provides a Euclidean inner

2.2. Source space. Let Σ be a smooth closed and oriented surface,

2.3. Fiberwise structures on tensor bundles. The bundle T ∗ Σ → Σ

We consider the Riemannian metric g given by the real part of h.

g(iF1 , iF2 ) = g(J · F1 , J · F2 ) = g(F1 , F2 ).

The formal adjoint d⋆ of d is deﬁned by the property

the C k,ν -norm of F control the C k−2,ν -norm of ∆2 G. The operator ∆

where ω(Ḟ1 , Ḟ2 ) = g(J · Ḟ1 , Ḟ2 ).

2.8. Gauge group action. We consider the inﬁnite dimensional com-

2.9. Infinitesimal gauge group action. For ζ ∈ tC = C ∞ (Σ, C) we

Formula (2.6) shows that

Lemma 2.10.1. The following formulas hold

Proof. By deﬁnition of φ(F ) = 12 kµ(F )k2 hence

3. The modified moment map flow

We deﬁne the modified moment map flow by the evolution equation

3.1. Basic properties of the flow. An interesting feature of the ﬂow is

3.2. Energy as a Morse-Bott function. By Proposition 2.13.1, we have

Theorem 3.2.2. We assume that dimR V = 4 and let f : Σ # V be

4. Polyhedral isotropic surfaces

Let gΣ = (gσ )σ∈K , be a family, where gσ is a Euclidean inner product

4.2. Whitney cohomology. The theory of Whitney forms and Whitney

which deﬁnes the Whitney cohomology denoted Hwn (Σ, R, T ). The

4.3. Orientation and Kähler structure. An orientation of the simpli-

of Σ. An oriented polyhedral surface carries a canonical area form

By construction, df is a Whitney form df ∈ Ω1w (Σ, V~ , T ). The moduli

These observations motivate the introduction of a space of constant

The integrands of the above integrals are constant on each simplex.

An almost complex structure J on F (T ) is given by

4.6. Polyhedral symplectic density. We denote denote by tC (T ) the

Definition 4.6.2. A differential F ∈ F (T ) is called isotropic if Fσ :

In particular, by Lemma (4.6.1), we have

Lemma 4.6.3. A differential F ∈ F (T ) is isotropic if, and only if,

4.7. Hamiltonian gauge group action. We deﬁne the complex gauge

where F ± are the component of F given by the splitting F (T ) =

Proposition 4.7.1. The TC (T )-action on F (T ) perserves the almost

Dhhµ, ζii = −ιXζ Ω.

Proof. The proof is formally identical to the smooth setting.

Proposition 4.8.3 has the following strong consequence:

Corollary 4.8.4. For every F ∈ F0 (T ), there exists a unique solution

Proof. The short time existence is immediate Proposition 4.8.3 and

4.9. A Duistermaat theorem. In fact, the polyhedral modiﬁed moment

Theorem 4.9.1. For every F ∈ F0 (T ), the solution Ft of the modified